+ =
D f f 7 . 3 Re
51 . 2
log 2
1
Eq. 1
Experiments have however shown that different surfaces give different friction factor characteristics.
This is particularly valid for the transition region, where the flow changes from smooth turbulent to fully rough
turbulent. None has succeeded in explaining why the different surfaces give exactly the friction factor
characteristics they give, or predicting friction factor based on measurements of the physical wall surface.
Even if accepting the ColebrookWhite correlation as the valid one, it remains to find the hydraulic
roughness (). This can differ significantly from the physically measured wall roughness. Research has
proposed to set hydraulic roughness equal 1.55 times the measured wall roughness (see e.g.
Langelandsvik et al. (2008) and Shockling et al. (2006)). The physical roughness height in commercial steel
pipes is very low. Gassco has measured it to be in the range of 2 to 5 micron. When scanned by a human
finger most observers would characterize this as perfectly smooth. However, at high enough Reynolds
numbers, the laminar sublayer next to the wall diminishes making the roughness elements protrude through
this layer and into the turbulence and adding resistance, which is what defines the transition region.
The uncertainty associated with a priori calculation of wall friction and pressure drop in a pipeline
makes it necessary to have the friction or roughness tuned in a fullscale test in either way. The methods
employed by Gassco are described below.
b. Capacity Test
When the pipeline is installed more accurate aslaid data with respect to length, wall layer
thicknesses and burial depths are available and these data are used to update the computer simulator.
Shortly after startup a socalled capacity test is performed. Particular care is taken by all supply and delivery
points to operate the pipeline very steady for a period of 15 days. The best period of approximately 12 hours
duration is chosen as the official test period, and assumed to represent a steadystate condition in the
pipeline.
Singleleg
For a singleleg pipeline the hydraulic roughness is mainly a function of the following parameters
( ) C T T Q P P f
ambient in out in
, , , , , =
Eq. 2
where P denotes the pressure, Q is the standard volumetric flow rate, T is the temperature and C is
the gas composition.
The averaged boundary pressures from the test period are used as boundary conditions in a steady
state simulation. The hydraulic roughness can be determined through iterative model simulations where the
roughness is adjusted until the simulated flow rate equals the weighted average of the measured flow rates
from the test period. The resulting hydraulic roughness is then said to be this pipelines hydraulic roughness.
The procedure is illustrated in the figure below.
Figure 2 Illustration of capacity test methodology.
The other parameter that can be used to check if the model is a good representation of the physical
pipeline is the simulated outlet temperature versus the measured one. A deviation can have three different
causes. First is obviously the ambient temperature used in the model. If this deviates from the actual
temperature at the time of the test, it will usually result in a different modeled temperature. Second are the
pipeline parameters like wall data and burial depth. The modeled heat transfer and subsequently the
temperature will be affected if these parameters are incorrect in the model. Last come the equations in the
simulator software. If they fail to model the heat transfer between the surroundings and the gas, the joule
thompson cooling or the frictional heating effect correctly, the temperature will be affected. And an inaccurate
simulated gas temperature along (parts) of the pipeline will affect the calculated hydraulic roughness. Care
must therefore be taken to reduce the possible deviation between simulated and measured outlet
temperature. Gassco has checked the equations in the simulator, so focus is on the two first causes when
we see a temperature deviation in a capacity test.
The capacity test is often performed at flow rates significantly lower than the maximum capacity, due
to limited amount of gas available early in a pipelines lifetime. The simulator is hence used to calculate the
hydraulic capacity by using maximum inlet pressure and minimum outlet pressure. This implies extrapolating
Ambient temperature
Elevation
Length
Diameter
Simulator
ColebrookWhite
P
out
P
in
T
in
Composition
in
Measurements
Simulations
Flow rate
Flow rate
Roughness
Measurements
Model
the friction factor along the specific ColebrookWhite line for this roughness in the Moody diagram to find the
friction factor at maximum capacity. It hence relies on the accuracy of the ColebrookWhite correlation, and
will be further described in a later section.
Network with several supply and delivery points
The method for estimating the effective roughness for a pipeline network is more complicated than
for a single pipeline. Figure 3 shows a schematic example of a pipeline network which consists of a main
pipeline and three branches. The tiein points are denoted A, B and C. A compressor, K, is also represented.
The main pipeline and the branches may have different physical properties like diameter, physical roughness
etc.
Figure 3 Example of a pipeline network.
As shown in Figure, the network can be considered as a connection of the following singleleg
pipelines:
Table 1: Network elements in the example network of Figure 3.
From To Color in Figure 3
Inlet1 A Pink
Inlet2 A Blue
A B Red
Inlet3 B Black
B K Yellow
K C Gray
C Outlet1 Green
C Outlet2 Brown
If measurements are available at the tiein points A and B as well as upstream and downstream the
compressor, an effective roughness can be determined for each singleleg that constitutes the network.
The methodology is then similar to the one described above.
When there are no pressure instruments at the tiein points, the singleleg tuning approach is
impossible. The following parameters are then necessary in order to determine the effective roughness in the
different parts of the network:
Pressure at the network inlets: P
in,i
, where i = inlet number 1,2,,n
in
Pressure at the outlets: P
out,j
where j = outlet number 1,2,,n
out
Flow rate at the inlets: Q
in,i
Flow rate at the outlets: Q
out,j
Temperature at the inlets: T
in,i
K Inlet 1
Inlet 2
Inlet 3
Outlet 1
Outlet2
A B
C
K 1
2
3
1
2
A B
C
Ambient temperature along the pipeline: T
ambient
(x,y)
Composition at the inlets: C
in,i
In other words,
) , , , , , , (
, , , , , , i in ambient i in j out i in j out i in
C T T P P Q Q f =
Eq. 3
The boundary conditions that are recommended to use in the model simulations when the effective
roughness is adjusted are shown in Table 2.
Table 2: Boundary conditions to use in the simulations.
Description Symbol
Flow rate at the inlets Q
in,i
Outlet pressures P
out,j
Ambient temperature T
ambient
Temperature at the inlets T
in,i
Gas composition at the inlets C
in,i
To achieve a steady state solution, the model needs either pressure or flow at each inlet and outlet,
where the pressure has to be provided for at least one inlet/outlet. The other pressure and flow
measurements are redundant. The boundary conditions selected in Table 2 represent what is usually chosen
in a capacity study. Other boundary conditions would have worked equally well. Each redundant
measurement can be used to tune one parameter, usually one effective roughness.
To match the test data measurements from the capacity test, it is necessary to perform iterative
simulations to find the matching simulation parameters shown in Table 3.
Table 3: Simulation output parameters.
Description Symbol
Pressure at all inlets P
in,i
Flow at (n
out
1) outlets F
out,j
The flow at the last outlet is determined by the flow at the other inlets and outlets, keeping in mind
that at steadystate the total inlet flow must equal the total outlet flow.
Lack of knowledge on the real system (burial depth, ambient temperature, material etc.) might result
in poor outlet temperature predictions. Nevertheless, the simulated outlet temperature must be checked
against specified pipeline temperatures to ensure that the simulated gas transport scenarios give acceptable
results.
For pipelines without compressors, the effective roughness of the leg considered to be the main
pipeline in the network is found when the simulated and measured pressures at the main pipeline inlet are
the same. This leg should be tuned before starting with the branches. In Figure 3, the main pipeline will be
the pipeline running from Inlet 1 to Outlet 1.
When the effective roughness is determined for the main pipeline, the simulation also gives the
pressures at the tiein points. Possible deviations between the simulated and the real tiein point pressures
(which are not measured) will obviously not be detected.
To match the test data measurements at the other branches, it is necessary to perform iterative
simulations for each branch to match the simulated and measured inlet pressures.
Two methods of matching the pressure measurements at the branches are given in prioritized order
here:
1. Tune separate effective roughness for each branch connected to the main pipeline.
2. Use a chosen effective roughness for each branch and tune a flow resistance element at the
end of each branch. The modelled resistance coefficient must be tuned to match the desired
pressure drop.
For both alternatives an iteration process is necessary to match the measurements. In case of
relatively short branches with low flow rates, the adoption of the second approach may become necessary
since very large changes in roughness are needed to achieve the desired flow resistance in the branch
pipeline at low flow rates.
Often the most important aim of a capacity test is to estimate the roughness of the main pipeline, and
subsequently find the capacity of the main pipeline, i.e., the output capacity.
Steadystate flow weighing
In a steadystate simulation, the sum of flow into the network needs to equal the sum of the flow out
of the network. This is not necessarily the case for the real pipeline during the capacity test, either due to
small remaining transients or due to metering errors. The flow rate that shall be obtained in the steadystate
simulation is calculated by weighing the different metered flow rates. For a singleleg pipeline this means
flow rate is calculated by:
out in mean
Q w Q w Q + = ) 1 (
Eq. 4
where the weight w is calculated based on the flow meters uncertainties to minimize the uncertainty
in Q
mean
. It can be shown that this is obtained by selecting:
2 2
2
out in
out
u u
u
w
+
= Eq. 5
where u denotes the uncertainty.
For a pipeline network with several supply points and/or several delivery points the calculation is
similar, though a bit more complex.
c. Operational Data
The two major drawbacks with the capacity test approach described above are that it relies on one
testing point and that it often is performed at a low flow rate and therefore relies on extrapolation of the
friction factor along a ColebrookWhite curve. The approach described here adds useful knowledge about
the pipeline and its capacity in addition to that obtained from a capacity test.
Uncertainty analysis quantifies the capacity calculation uncertainty from a capacity test. This
uncertainty comprises both systematic error and random error. The categorization of systematic and random
error is often difficult, and one usually only obtain a vague idea of their relative contribution to the total
uncertainty figure. It is well known that the random error can be minimized and eventually be made neglible if
more testing points are averaged to find a pipelines hydraulic roughness. Gasscos capacity calculation
methodology has therefore been extended to average a set of steadystate period data points to calculate
the hydraulic roughness. And instead of organizing an elaborate capacity test to obtain every single data
point, a data base of historical operational data for the pipeline is used. This data base contains logged data
from all pressure and temperature transmitters, gas chromatographs, flow meters and other instruments
connected to the pipeline. All data are usually logged with intervals of approximately 1 minute. This data
base is searched to find good steadystate operational periods which have occurred arbitrarily in the daily
operation of the pipeline. A certain set of criteria has been developed for the periods to qualify as a steady
state period. Each of these periods is then treated exactly like a capacity test period, and have a roughness
tuned.
The ColebrookWhite friction factor correlation has been accepted as an industry standard for
decades, even though many research groups have proved it to be wrong in their specific tests. The reason it
is still well accepted is that none has succeeded in explaining why the ColebrookWhite correlation fit or does
not fit to their experiments and how different wall surface structures lead to different transition regions (see
e.g. results from American Gas Association in the 1960s in Uhl et al. (1965) and more recent experiments in
Superpipe at Princeton University by Shockling et al. (2006)). The uncertainty associated with extrapolating
along a ColebrookWhite curve can be almost entirely removed by selecting steadystate periods with high
flow rates. Therefore only steadystate periods with a flow rate of more than approximately 80% of the
pipelines expected capacity are used when averaging the hydraulic roughness. Most of the Gassco
operated pipelines are in the early phase of the transition region from smooth turbulent flow to fully rough
turbulent flow, where ColebrookWhite is mostly questioned. Evaluating only steadystate periods with high
flow rates makes this uncertainty neglible.
Figure 4 shows steadystate periods that have been collected and simulated for one specific export
pipeline, denoted pipeline A. We see that all the data points have a roughness in the range of 1.5 to 3.0
micron. The average roughness of the data points with highest Reynolds number seems to be approximately
2.2 micron. Imagine the encircled data point which have a roughness close to 3 micron was a capacitytest
data point. Extrapolating along the corresponding ColebrookWhite curve would have yielded a too large
friction factor (marked with red unfilled circle) at maximum capacity and thus predicting a too low capacity.
The red filled circle illustrates the recommended friction factor based on averaging steadystate operational
data periods. Vice verca would a low capacity test result over predict capacity and result in over booking.
This illustrates the reduced uncertainty that is gained by averaging a set of highflow data points, even
though this example shows data points that do not deviate a lot from the ColebrookWhite trends.
One can also obtain an experimental friction factor curve for the pipeline in question by fitting a line
to the plotted friction factors from the steadystate periods. For a limited range of Reynolds number, e.g.
between 10 and 4010
6
, which is the operating range for pipeline A and most others Gassco operated
pipelines, it is usually sufficient to use a straight line. Regression analysis can be used to find the line. For
pipeline A such a line would be quite parallel to the ColebrookWhite lines. If the steadystate periods do not
cover flow rates close to the maximum flow rate, even extrapolation along a fitted pipeline specific friction
factor curve would imply uncertainty, since one cannot predict the behaviour of the friction factor for larger
flow rates.
The physical roughness for pipeline 1 has not been measured. But based on measurements on other
pipelines which have been manufactured according to the same specifications, it is believed that the root
mean square roughness is around 23 micron. This is about the same value as the hydraulic roughness
estimated to be 2.2 micron in this case. This unexpected low hydraulic roughness may indicate that the
friction factor characteristics will deviate from the ColebrookWhite lines for larger Reynolds numbers. This is
discussed in more detail in Langelandsvik (2008).
Friction factor results, Zeepipe
6.00E03
6.50E03
7.00E03
7.50E03
8.00E03
8.50E03
9.00E03
10 000 000 100 000 000
Reynolds number, Re []
f
r
i
c
t
i
o
n
f
a
c
t
o
r
[

]
0.01 micron 1.3 micron
2 micron 3 micron
5 micron test points
Figure 4 Moodydiagram where the lines are given by the Colebrookwhite correlation for different relative
roughnesses. Data points from simulation of steadystate periods for the export pipeline A.
In a capacity test, the instruments are usually calibrated beforehand, and special effort is taken by
the field and plant operators to ensure steadystate conditions in the pipeline during the test. In operational
steadystate periods none of these keys to success are present. But despite this, the benefits of the new
approach more than outweigh these drawbacks.
The increased accuracy and reduced margins have lead to an increase in calculated transport
capacity of in total 4.6 MSm3/d for five export pipelines. It should be noted that one pipeline contributes 2.7
MSm
3
/d to this number. The basis for this pipelines capacity was design capacity and not capacity test
capacity as for the other pipelines, ie. the capacity was even more conservative before the application of the
described methodology.
Requirements for steadystate periods
It is important that the operational period is a good steadystate representation of the pipeline for the
given flow rate. The steadystate periods are therefore found in a twostep procedure. The database that
contains the operational data provides a powerful search tool, which can suggest a set of steadystate
periods based on criteria set by the user. But every single period is also checked visually by Gasscos
Possible capacity test
point giving a
conservative capacity
Recommended friction
at maximum capacity
based on steadystate
operational data
Recommended friction
at maximum capacity
based on possible
capacity test point
Friction factor results, pipeline A
engineers to reveal if any unstability is present before or in the period.The set of steadystate criteria that has
been developed is shown below.
Table 4 Criteria for steadystate periods.
Topic Requirements
Absence of
transients
12 hours prior to the test period
Period of
steadiness
Preferably 24 hours, but at least 12 hours
Spacing
Two steadystate periods should not be too close in time. If the pressures or flow
rates have changed it is ok to have a short time spacing, otherwise at least 1
week shall be used.
Flow
Average growth rate less than 5% of the flow rate per day.
Standard deviation of no more than 2 % of the average value
No step in the trend
Temperature As stable as possible, at least no large transients should be present
Pressure
Average growth rate less than 3 barg/d
Standard deviation of no more than 2 % of the average value
No step in the trend
Composition Stability within one standard deviation (uncertainty for the gas chromatograph).
Packing A maximum average rate of packing or depacking of 0.2 MSm
3
/d
The criteria are based partly on experience, and partly on results from a certain pipeline which
showed that the variation in hydraulic roughness suddenly increased when including slightly poorer steady
state periods.
4 Ambient Temperature
The importance of simulating a correct gas temperature profile along the pipeline in order to
calculate the correct hydraulic roughness was slightly touched upon above. And having a good estimate of
the ambient temperature at the time of the steadystate period is clearly a condition in simulating a correct
temperature profile. But getting a correct hydraulic roughness for the pipeline is only halfway in the capacity
calculation. To utilize this information one also needs to know the expected ambient temperature at the time
the calculated capacity shall be valid for.
Nearrealtime seawater temperatures along pipelines may be obtained from the use of numerical
ocean models. A total of more than 50 ocean circulation models are presented on the Internet. Some of
these models are freely available, some are freely available for research purposes only and some are not
available at all (or available at a cost). Some model operators (notably meteorological institutes) make data
from their model runs commercially available. Typically the data are provided as nowcasts and up to five
day forecasts.
Of the operational numerical models that cover the North Sea only the UK Met Office Shelfseas
model was considered to provide sufficiently accurate sea bottom temperature data for Gasscos needs.
Comparison of measured and modelled data concluded that the average total error between model and
measured data is slightly above 1 C. The Shelfseas model performs best in the southern North Sea where
the water depths are less than 50 m and the temperature fluctuations are the largest. See Hendriks et al.
(2006).
A link between the Shelfseas model and Gasscos real time simulation software was implemented in
2006. This provides daily and direct access for the real time simulation software to nowcasts and twoday
forecasts of the sea bottom temperature along all export pipelines. At the same time Gassco obtained
access to historically calculated nowcasts for every single day since the Shelfseas model was started in
2002.
The historically calculated nowcasts are now used as input when simulating every historical
operational steadystate period. This gives a better estimate of the actual seabottom temperatures during
each steadystate period than the previously used tabulated statistical data. It hence represents an
improvement with respect to ambient temperature uncertainty, and gives a more accurate calculation of the
hydraulic roughness for the pipelines.
In Figure 5 the simulated roughness values for a set of steadystate periods are shown. After
implementation of modelled data from the Shelfseas model, it is seen that the variation reduces
considerably. In this case the standard deviation is reduced from 0.49 m to 0.36 m. (26%).
0.00E+00
1.00E06
2.00E06
3.00E06
4.00E06
5.00E06
20011105 20020124 20020414 20020703 20020921 20021210 20030228 20030519
Date
E
f
f
e
c
t
i
v
e
R
o
u
g
n
e
s
s
[
m
]
Table averages
Daily grids
Figure 5: Calculated effective roughnesses in pipeline A, based on operational data from steadystate periods.
Blue diamonds represent uptodate daily predictions for ambient temperature and pink squares are based on
old monthly tabulated averages.
In addition to the Shelfseas model, historical averages from the World Ocean Atlas 2005 (WOA
2005) are compiled in grid files for use in the simulation software. One dataset is given for each month and is
valid for the middle of that month. Data for an arbitrary date is obtained by interpolating between the two
nearest months. These historical temperatures are the best available predictions of the future. A
conservative base case capacity is calculated for the pipelines using the roughness and employing WOA
2005 data for the warmest month in the year, usually October. This capacity is sold in booking rounds two
times a year on longterm, currently meaning up to 2028. In the same booking rounds, the shippers can book
monthly capacities on medium term which is two years ahead. On mediumterm one takes advantage of the
seasonal temperature variation in the sea, which increases the capacity in the cold winter months when
natural gas demand and transport capacity demand peaks. In addition the operational flexibility is often
reduced from 2% to 1% on medium term. Shortterm booking happens every Friday, where available and
additional capacity is offered on a daily basis five weeks ahead. Additional margins are exploited here.
Gassco plans to use the twoday temperature forecasts received from the Shelfseas model to take
advantage of the daily variations in temperature and capacity, and include this in the shortterm booking.
Additional capacity due to temperature effects included in mediumterm and shortterm booking amounts to
35 MSm
3
/d during winter. This temperature effect is well illustrated in Figure 6.
41.5
42.0
42.5
43.0
43.5
44.0
1 2 3 4 5 6 7 8 9 10 11 12
Month
F
l
o
w
r
a
t
e
[
M
S
m
/
d
]
Low ambient temperature
Average ambient temperature
High ambient temperature
Capacity offered for sale [old situation]
Figure 6 Example hydraulic capacities for a pipeline influence by ambient temperature.
5 Heat Transfer for Partly Buried Pipes
Most of the Gassco operated subsea pipelines are partly buried on the seabottom. Originally they
are either laid directly on the seabottom such that seacurrents cause them to be partly buried in sand or
clay after a while, or they are trenched in order to achieve the desired stability. In both cases the heat
transfer takes place partly through the ground by conduction, and partly through the seawater by conduction
and natural and forced convection.
Gersten et al. (2001) concludes that heat transfer to/from entirely buried pipes is well known,
provided the characteristic of the ground is known. Heat transfer to/from pipes entirely exposed to sea water
is also fairly well modelled. But none has focused on the intermediate situation where a combination of these
two effects takes place. Gersten et al. suggested a linear interpolation between these two effects. Later
Morud et al. (2007) presented an analytical model for this combined situation, and Morud et al. and Ramsen
et al. (2009) showed that the Morud model fit well with CFD simulations for different burial depths.
More detailed heat transfer coefficients obtained from this model have been used by Gassco for one
specific pipeline. Information about the burial depths from a recent survey of the entire pipeline was used in
the calculation of the coefficients. The capacity calculation of this pipeline has been checked based on this
updated model, and provides an example of how latest knowledge about all aspects of fluid mechanics shall
be used to improve the accuracy and reduce the margins in transport capacity calculation.
6 Uncertainty Calculations
The uncertainty of the calculated roughness for a capacity test period or a steadystate operational
period is calculated according to an established scheme.
First all parameters affecting the calculated roughness are identified. This includes metering signals
such as pressure, flow rate, gas temperature and gas composition and the ambient temperature estimate.
For the metering signals the standard deviation, , is calculated, which is an estimate of the parameters
repeatability. This is combined with the instruments stated uncertainty to find the total uncertainty of the
input parameter, )
(
i
X u .
The approach assumes that the roughness calculation is linear with respect to all input parameters,
which is a viable assumption close to the starting point. Furthermore it utilizes the fact that the variance of a
linear combination of stochastic variables, Z = aA+bB is given by:
) , ( 2 ) ( ) ( ) (
2 2
B A abCov B Var b A Var a Z Var + + =
Eq. 6
On a cold day
1.2 MSm/d extra can
be made available for
this pipeline
It is assumed that the input parameters uncertainty is uncorrelated, and the covariance term is
hence disregarded. For each of the parameters, the hydraulic roughness is calculated when the current
parameter take the value )
(
i i i
X u x x = , while the other parameters ( i) takes the average value. The
total uncertainty of the calculated hydraulic roughness, ) ( u , is then calculated by combining the
uncertainty contributions from the different parameters which can be simplified to:
( )
=
+
n
i
n i i n i i
x X u x x f x X u x x f u
1
2
1 1
2
) ),...,
( ,..., ( ) ),...,
( ,..., (
4
1
) ( Eq. 7
Typical uncertainty in the roughness calculation for a capacity test is around 0.5 m, which often
means an uncertainty in capacity of around 0.5 MSm
3
/d. For a steadystate operational period it is slightly
higher due to the imperfect steadiness. The flow meter uncertainty is the largest contributor to the
uncertainty.
The uncertainty is comprised of random error (precision) and systematic error (accuracy) according
to:
2 2 2
) ( ) ( ) ( accuracy precision u +
Eq. 8
When the estimation procedure implies averaging roughness across many periods, it can be shown
that the uncertainty contribution from precision decreases and eventually approaches zero.
What remains is the accuracy, which is a systematic error affecting each period equally. This cannot
be reduced by adding more steadystate periods. The standard deviation of the different roughness
calculations can be used as an estimate of the precision.
7 Summary and Perspectives
We have here shown how the transport capacity of both singleleg pipelines and networks may be
calculated in a very accurate manner. Following a well planned steadystate capacity test, systematic
analysis of operational data with high flowrates, implementation and utilization of uptodate modelled
ambient temperature and systematic improvement of all parts of the model equations and correlations all
contribute to this. It is also shown that this increased knowledge about the fluid dynamics in the pipelines has
revealed that the actual physical capacity is higher than previously committed, which obviously is of great
importance to all shippers and operators in the North Sea. The increased knowledge has resulted in an
increase in committable capacity of 4.6 MSm
3
/d on average with an additional 35 MSm3/d that can be
exploited on cold winter days.
Gassco is currently supporting a work to investigate the potential for improvement in the Shelfseas
model. In particular one has seen that the deviation between measured and modelled outlet temperatures
strongly varies with season. The model predicted gas temperature is typically too low in the winter and too
high in the summer. Whether this is caused by the Shelfseas model or other sources of uncertainty remains
to be analyzed.
The capacity calculation methodology can still be improved in some ways. The uncertainty analysis
for the steadystate operational data can be improved. The uncertainty contribution from the slight decrease
in steadiness in a period should be subject to further analyses.
References
[1] Gersten, K., Papenfuss, H.D., Kurschat, T.H., Genillon, P.H., Fernandez Perez, F., Revell, N.
(2001). Heat Transfer in Gas Pipelines, OIL GAS European Magazine, 1/2001.
[2] Hendriks, P.H.G.M., Postvoll, W., Mathiesen, M., Spiers, R.P., Siddorn, J. (2006). Improved
Capacity Utilization by Integrating Realtime Sea Bottom Temperature Data. Proceedings of the 2006
PSIG Conference, Williamsburg, Virginia, US.
[3] Langelandsvik, L.I., Kunkel, G.J., Smits, A.J.(2008). Flow in a commercial steel pipe. Journal of
Fluid Mechanics, Vol. 595, pp. 323339.
[4] Langelandsvik, L.I. (2008). Modeling of natural gas transport and friction factor for large scale
pipelines Laboratory experiments and analysis of operational data. PhD dissertation, Norwegian
University of Science and Technology, 2008.
[5] Morud, J.C., Simonsen, A. (2007). Heat transfer from partially buried pipes. Proceedings of the 16
th
Australasian Fluid Mechanics Conference, Crown Plaza, Gold Coast, Australia, 2007.
[6] Ramsen, J., Losnegrd, S.E., Langelandsvik, L.I., Simonsen, A., Postvoll, W. (2009). Important
Aspects of Gas Temperature Modelling. Proceedings of the 2009 PSIG Conference, Galveston Texas,
US.
[7] Shockling, M. A., Allen, J. J., Smits, A. J. (2006). Roughness effects in turbulent pipe flow. Journal
of Fluid Mechanics, Vol. 564, pp. 267285.
[8] Uhl, A., Bischoff, K.B., Bukacek, R.F., Burket, P.V., Ellington, R.T., Kniebes, D.V., Staats, W.R.,
Worcester, D.A. (1965). Steady flow in gas pipelines. Institute of Gas Technology Technical report no.
10. American Gas Association.